Efficient Quantum Control via Automatic Control Skips

Eyal Cornfeld; Israel Reichental; Peleg Emanuel; Ravid Alon; Shmuel Ur

REVIEW 3 major objections 5 minor 29 references

Efficient Quantum Control via Automatic Control Skips

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Controlled quantum gates can be automatically skipped in generic circuit decompositions, with over 50 percent savings on state-preparation benchmarks.

desk verdict Useful compiler-pass idea with a load-bearing DP typo that needs fixing; the MCP formalization and NP-hardness proof are the real contributions. read the letter →

arxiv 2505.18256 v1 pith:JZUF5627 submitted 2025-05-23 quant-ph

Peleg Emanuel , Eyal Cornfeld , Ravid Alon , Shmuel Ur , Israel Reichental This is my paper

classification quant-ph MSC 81P6868Q1768Q25

keywords quantumcontrolcontrolledoperationsconjugationpairscircuitoptimizationMaxNP-hardnessdynamicprogrammingcompilation

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper argues that the high cost of controlled quantum operations can be attacked automatically, without algorithm-specific manual design. It formalizes the task as Max Conjugation Pairs (MCP): find a maximum-weight set of inverse-operation pairs such that omitting their controls still implements the same unitary. The paper proves MCP is NP-hard by reduction from maximum-weight independent set, then gives a polynomial dynamic-programming approximation for the non-commuting case. On state-preparation circuits the method reduces gate counts and depth by over 50 percent, while on random brickwork circuits it gives roughly 8 percent. The point is that a compiler pass can specialize controlled circuits without knowing what the algorithm does.

What carries the argument

The load-bearing object is the conjugation pair: a pair of inverse unitaries $U_i$ and $U_j = U_i^{-1}$ in the decomposition, recorded with a weight. A collection of such pairs is skippable when any two pairs commute or their index intervals do not interlace; in that case the product over the selected operations is the identity, and the identity in Eq. (2) licenses removing their controls. The argument then reduces the non-commuting case to a dynamic-programming recurrence, Eq. (3), which splits the operator sequence at a chosen pair and combines weights from disjoint ranges. Solving that recurrence over all ranges gives the reported $O(r n^2)$ runtime, with $r$ the maximal multiplicity of an operator.

What would settle it

Apply the printed Eq. (3) to the four-operator sequence $A, A, A^{-1}, A^{-1}$. If the algorithm chooses the first pair $(1,3)$, the inner subproblem becomes range $[2,3]$, which can also choose the pair $(2,3)$; the resulting selected set multiplies to $A$ instead of the identity, so a correct implementation must instead return exactly one pair.

Watch

Extended reading notes

Core claim

The central claim is that whenever a product of some operations in a decomposition equals the identity, the control of those operations can be dropped from the controlled version of the whole unitary. The paper defines the weighted Max Conjugation Pairs problem: choose conjugation pairs $(U_i, U_i^{-1})$ with maximum total weight, subject to a compatibility condition that any two chosen pairs either commute or do not interlace. It proves that finding the optimal choice is NP-hard, and it claims that a sub-optimal polynomial algorithm based on a dynamic-programming recurrence over non-commuting ranges still returns a valid set of skippable operations. In real-world state-preparation circuits this lowers gate counts and depth by more than half, and the method is intended to sit on top of hand-written specialized implementations.

Load-bearing premise

The whole method rests on the dynamic-programming recurrence in Eq. (3) exactly characterizing the non-commuting problem; as printed, its inner subproblem range would allow the same matched gate to be used twice, so the paper's guarantee depends on that recurrence being corrected as described in the text.

Editorial extensions

If this is right

If the claim is correct, a compiler can automatically detect control-skipping opportunities in arbitrary decompositions, at any description level, without programmer effort.
Because MCP reduces to Maximum Weight Independent Set, existing MWIS approximation algorithms and practical solvers can be applied directly to find good skippable sets.
The method composes with hand-written optimizations: a programmer can fix certain conjugation pairs and let the algorithm complete the rest of the set.
Lowering two-qubit-gate counts and depth in controlled operations should reduce the cost of fault-tolerant implementations, since controlled gates are a major source of overhead.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A natural next step would be to look for identity subproducts beyond pairs, since the underlying condition in Eq. (2) is not limited to conjugation pairs.
The paper leaves open the choice of basis decomposition; experiments that vary which single-qubit gates appear in the basis could expose new conjugation pairs in otherwise opaque sequences.
The non-commuting DP is applied even when real circuits contain commuting blocks; scheduling or rewriting circuits to enlarge commuting regions could improve the approximation's performance.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The paper is worth reading because it turns a clever but ad hoc quantum-compiler trick into a clean combinatorial problem. The formalization of Max Conjugation Pairs (MCP), the NP-hardness reduction from MWIS, and the geometric DP for the non-commuting case are genuinely new. The core identity (skip any subcircuit that multiplies to identity) is standard, but the systematic treatment as a weighted selection problem is a real contribution that could slot into a compiler pipeline.

That said, the paper has one load-bearing flaw in the text. Equation (3) recurses on W_{i+1,k} in the branch where operation i is paired with k, but the accompanying text and the logic of non-overlapping intervals require W_{i+1,k-1}. As printed, the recurrence can reuse operation k in a second pair, and the claimed guarantee that the algorithm 'still outputs a valid solution to the general problem' is false. This is almost certainly a typo, since the text describes the correct subranges, but no code is released to confirm the implementation matches the text rather than the equation. The empirical headline—over 50% reduction on state preparation—depends on this recurrence producing valid skips, so this has to be fixed and ideally verified with released code.

Other soft spots are proportionately minor. The novelty relative to the authors' own 2024 report (Ref [18], 'Automatic quantum circuit control skips') is not clearly delineated; the paper should state which results are new. The experiments use hand-picked weights (w(CX)=3, others=1) and report no error bars or statistical variation across the five instances per error bound, so the over-50% claim is plausible but not tightly quantified. The finite-basis-gate extension in the NP-hardness proof is also loosely stated: the h_{kl}(θ) family is continuous and parameterized, so the 'nine basis gates' comment needs unpacking. None of these undermine the core argument after the recurrence fix.

Overall: a solid, useful paper that needs a careful revision before it can be trusted as published. The MCP formulation and the reduction are worth a serious referee even though the empirical section is a bit thin. I would want to see the corrected recurrence, ideally with code, and a clearer novelty statement relative to the earlier report.

Referee Report

3 major / 5 minor

Summary. The paper introduces the Max Conjugation Pairs (MCP) problem: given a unitary decomposition, find a maximum-weight set of conjugation pairs (U_i, U_i^+) that can be left uncontrolled in a controlled circuit, with validity verified by requiring that pairs either commute or do not interlace. It proves MCP NP-hard by reduction from Maximum Weight Independent Set, presents a polynomial-time dynamic program (Eq. (3)) for the restricted non-commuting case, and reports experiments on state-preparation and brickwork circuits showing up to 50% reductions in gate counts and depth. The central claim is that this is a generic, automatic compiler-level optimization that does not require algorithm-specific tailoring.

Significance. If the claims hold, the paper offers a useful compiler pass for optimizing controlled quantum circuits. The NP-hardness reduction is explicit and self-contained, and the dynamic program is simple and polynomial. The empirical demonstration, while limited to a naive baseline and a small number of circuits, suggests practical value. The authors correctly position the method as complementary to specialized control implementations. Because no code is released and one printed recurrence is inconsistent with the accompanying text, the current version does not yet provide a verifiable specification of the algorithm.

major comments (3)

[III, Eq. (3)] In the recurrence for W_{i,j}, the term for choosing the conjugation pair (i,k) is printed as W_{i+1,k} + W_{k+1,j}. Because the subproblem ranges are inclusive, W_{i+1,k} still contains index k, which is the operation U_k just matched with U_i. A second pair inside [i+1,k] could therefore reuse U_k, producing a selected set whose product is not the identity. The accompanying text states that the inner subproblem should be [2, j−1] (i.e., [i+1,k−1] in general). The printed equation should be corrected to W_{i+1,k−1} + W_{k+1,j}. As written, the algorithm is not guaranteed to output a valid solution, which undermines the reported improvements.
[II, Definition] The MCP definition says that two pairs can be chosen together if either the unitaries commute "or the sections [i(a),j(a)] and [i(b),j(b)] do not overlap." If "overlap" is read as having a common index, this forbids nested pairs (e.g., (1,4) and (2,3) in a sequence U1,U2,U2^+,U1^+), which are always skippable even when the unitaries do not commute. The subsequent geometric description and the DP in Sec. III assume the condition is "do not interlace" (non-crossing arcs), not "do not overlap." The authors should clarify or amend the definition so that the formal problem statement matches the algorithm and the earlier explanation.
[III, Fig. 4] The experimental section does not provide the implementation code or a pseudocode of the DP, the exact parameters for generating the random circuits, the number of brickwork instances, or error bars. Given the off-by-one issue in Eq. (3), a reader cannot verify that the reported improvements were produced by a correct implementation of the recurrence. The authors should release the code or provide a precise pseudocode and additional experimental details to make the headline "over 50% improvement" reproducible.

minor comments (5)

[I, Eq. (1)] The combined Hilbert space should be the tensor product H_control ⊗ H_target, not the direct sum H_control ⊕ H_target as written.
[II, Theorem proof] The reduction should choose θ_i values that are not integer multiples of π (e.g., θ_i=1) to ensure that e^{iθ_i g_i} and e^{iθ_j g_j} do not commute for anticommuting g_i,g_j; as stated, θ_i∈(0,2π) alone permits accidental identity operators.
[II, finite-basis extension] The claim that the NP-hardness proof holds with a finite set of parametric basis gates h_{kl}(θ) acting on consecutive qubits is under-specified: the operators U_i = exp(iθ_i ∏_{e incident to i} P_e) are rotations about Pauli strings supported on arbitrary subsets of edge-qubits, and it is not shown how these are expressed as products of the stated h_{kl} gates. Please expand this argument.
[III, Eq. (3)] The notation "k∈[i+1,j].U_k=U_i^{-1}" uses a period instead of a colon or comma; the condition should read "k∈[i+1,j] such that U_k=U_i^{-1}".
[III, Fig. 4] The state-preparation results are based on 5 instances per ϵ value but no standard deviations are shown, and the number of brickwork instances is not stated; please add error bars or box plots.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the control-skip identity, NP-hardness reduction, and DP algorithm are self-contained; self-citations to the authors' toolchain are not load-bearing.

full rationale

The derivation chain is self-contained. The enabling identity Eq. (2), ctrl(U) = product of controlled U_j with controls skipped on S where product_{j in S} U_j = 1, is proven in footnote [14] by direct multiplication for both control states; it is not assumed from the conclusion. The MCP problem is defined explicitly, and the NP-hardness proof is a standard reduction from Maximum Weighted Independent Set using operators U_i = exp(i theta_i g_i) whose commutation relations are verified from the Pauli X/Z terms; the proof does not invoke the target result. The dynamic-programming recurrence Eq. (3) is derived from the stated geometric 'non-interlacing' condition for non-commuting sequences. Even if the printed subrange W_{i+1,k} contains an off-by-one slip relative to the prose's [2, j-1], that is a potential implementation-correctness issue, not a circularity: an erroneous recurrence does not make the claimed derivation equivalent to its input. The empirical improvement is measured with explicitly stated hand-chosen weights w(CX)=3 and w(g != CX)=1, comparing the control-skipped circuit against the naively controlled circuit; no parameter is fitted to the reported gate-count or depth reductions, so the over-50% figure is a genuine difference between two explicit constructions rather than a renamed input. The self-citations ([17]-[19] and [18]) refer to the authors' Qmod compiler toolchain and an earlier report of the same method, but the mathematical claims and algorithms are reproduced in this paper and the cited works are not used as authority for correctness; hence they do not carry the derivation. Overall, no circular step reduces a stated prediction to an input by construction.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical entities. Its free parameters are limited to the gate cost weights. The key axioms are standard math plus a combinatorial lemma stated in the paper. The main concern is the unverified DP recurrence.

free parameters (1)

gate cost weights = w(CX)=3, w(g≠CX)=1
Hand-chosen cost model reflecting the higher cost of two-qubit gates; the algorithm's chosen skips depend on these values, but they are not fitted to the benchmark results.

assumptions (4)

standard math If a subset S of operations multiplies to the identity, control of those operations can be skipped (Eq. 2).
Stated in Section I and proved in footnote [14]; it is an elementary algebraic identity.
standard math A set of conjugation pairs in which every interlacing pair commutes has total product identity (footnote [20]).
This is the key combinatorial lemma justifying the MCP constraints; a proof sketch is given in footnote [20] but it is not formally verified.
standard math MWIS is NP-hard, including on degree-3 graphs (Refs [21,22]).
Used for the NP-hardness reduction.
domain assumption The non-commuting dynamic program applies to real circuits and still outputs a valid solution even though real circuits have commuting operations (Section III).
The DP only considers non-crossing arcs, which are always valid, but the paper does not prove this explicitly; the off-by-one issue in Eq. (3) casts doubt on the stated guarantee.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Efficient Quantum Control via Automatic Control Skips." pith.science (2026). https://pith.science/paper/JZUF5627

@misc{pith2026250518256,
  author       = {Pith},
  title        = {Pith review of: Efficient Quantum Control via Automatic Control Skips},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JZUF5627}},
  note         = {Machine review of arXiv:2505.18256}
}

abstract

Control of quantum operations is a crucial yet expensive construct for quantum computation. Efficient implementations of controlled operations often avoid applying control to certain subcircuits, which can significantly reduce the number of gates and overall circuit depth. However, these methods are specialized and circuits frequently need to be implemented manually. This paper presents a generic method for finding "skippable" patterns without having to tailor implementations for each algorithm. We prove that finding the optimal operations to be skipped is generally NP-hard. Nevertheless, sub-optimal, polynomial approximation algorithms that find skippable subcircuits can lead to over $50\%$ improvement in circuit metrics for real-world applications.

Figures

Figures reproduced from arXiv: 2505.18256 by the authors.

**Figure 2.** If one wishes to minimize the number of two-qubit [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗

**Figure 2.** FIG. 2: Circuits implementing the same controlled gate ctrl (1 [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

Figures from the paper (2 more)

**Figure 3.** Figure 3: FIG. 3: Illustration of conjugation pair choices for MCP. The pairs [PITH_FULL_IMAGE:figures/full_fig_p003_3.png]

**Figure 4.** Figure 4: FIG. 4: Circuit metric comparison of naively controlled circuits, and control-skip controlled circuits, using the non [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

29 extracted references · 13 canonical work pages

[18]

Naveh, S

A. Naveh, S. Ur, P. Emanuel, R. Alon, O. Kirzner, and E. Cornfeld, Automatic quantum circuit control skips (2024)

work page 2024
[1]

M. A. Nielsen and I. L. Chuang,Quantum computation and quantum information(Cambridge university press, 2010)

2010
[2]

N. P. de Leon, K. M. Itoh, D. Kim, K. K. Mehta, T. E. Northup, H. Paik, B. S. Palmer, N. Samarth, S. Sangtawesin, and D. W. Steuerman, Materials challenges and opportunities for quantum com- puting hardware, Science372, eabb2823 (2021), https://www.science.org/doi/pdf/10.1126/science.abb2823

doi:10.1126/science.abb2823 2021
[3]

Krantz, M

P. Krantz, M. Kjaergaard, F. Yan, T. P. Orlando, S. Gus- tavsson, and W. D. Oliver, A quantum engineer’s guide to superconducting qubits, Applied Physics Reviews6, 10.1063/1.5089550 (2019)

doi:10.1063/1.5089550 2019
[4]

C. D. Bruzewicz, J. Chiaverini, R. McConnell, and J. M. Sage, Trapped-ion quantum computing: Progress and challenges, Applied Physics Reviews6, 10.1063/1.5088164 (2019)

doi:10.1063/1.5088164 2019
[5]

Henriet, L

L. Henriet, L. Beguin, A. Signoles, T. Lahaye, A. Browaeys, G.-O. Reymond, and C. Jurczak, Quantum computing with neutral atoms, Quantum4, 327 (2020)

2020
[6]

Gidney, N

C. Gidney, N. Shutty, and C. Jones, Magic state culti- vation: growing t states as cheap as cnot gates (2024), arXiv:2409.17595 [quant-ph]

arXiv 2024
[7]

Maslov, Advantages of using relative-phase toffoli gates with an application to multiple control toffoli opti- mization, Phys

D. Maslov, Advantages of using relative-phase toffoli gates with an application to multiple control toffoli opti- mization, Phys. Rev. A93, 022311 (2016)

2016

Show all 29 references

[8]

Selinger, Quantum circuits of T-depth one, Phys

P. Selinger, Quantum circuits of T-depth one, Phys. Rev. A87, 042302 (2013)

2013
[9]

Huang and J

K. Huang and J. Palsberg, Compiling conditional quan- tum gates without using helper qubits, Proc. ACM Pro- gram. Lang.8, 10.1145/3656436 (2024)

2024 doi
[10]

E. C. R. Rosa, E. I. Duzzioni, and R. de Santiago, Op- timizing Gate Decomposition for High-Level Quantum Programming, Quantum9, 1659 (2025)

2025
[11]

R. Vale, T. M. D. Azevedo, I. C. S. Ara´ ujo, I. F. Araujo, and A. J. da Silva, Circuit decomposition of multicon- 6 trolled special unitary single-qubit gates, IEEE Transac- tions on Computer-Aided Design of Integrated Circuits and Systems43, 802 (2024)

2024
[12]

C. H. Bennett, Logical reversibility of computation, IBM Journal of Research and Development17, 525 (1973)

1973
[13]

S. A. Cuccaro, T. G. Draper, S. A. Kutin, and D. Petrie Moulton, A new quantum ripple-carry addition circuit, arXiv e-prints , quant-ph/0410184 (2004), arXiv:quant- ph/0410184 [quant-ph]

2004 arXiv
[14]

2 are identical regardless of the control state

We note that both sides of Eq. 2 are identical regardless of the control state. If the control state is 1, the left-hand side readsU, and the right-hand side reads Q j Uj, equal by the definition ofU. If the control state is 0, the left- hand side reads 1 and the right-hand si...
[15]

Yuan and M

C. Yuan and M. Carbin, The t-complexity costs of er- ror correction for control flow in quantum computation, Proc. ACM Program. Lang.8, 10.1145/3656397 (2024)

2024 doi
[16]

Beauregard, Circuit for shor’s algorithm using 2n+3 qubits, Quantum Info

S. Beauregard, Circuit for shor’s algorithm using 2n+3 qubits, Quantum Info. Comput.3, 175–185 (2003)

2003
[17]

Goldfriend, I

T. Goldfriend, I. Reichental, A. Naveh, L. Gazit, N. Yoran, R. Alon, S. Ur, S. Lahav, E. Cornfeld, A. Elazari, P. Emanuel, D. Harpaz, T. Michaeli, N. Erez, L. Preminger, R. Shapira, E. M. Garcell, O. Samimi, S. Kisch, G. Hallel, G. Kishony, V. van Wingerden, N. A. Rosenbloom, ...

2025 arXiv
[19]

M. Vax, P. Emanuel, E. Cornfeld, I. Reichental, O. Opher, O. Roth, T. Michaeli, L. Preminger, L. Gazit, A. Naveh, and Y. Naveh, Qmod: Expressive high-level quantum modeling (2025), arXiv:2502.19368 [quant-ph]

2025 arXiv
[20]

LetSbe a set of conjugation pairs such that if two pairs interlace, they commute

This may be proved by induction over the number of interlacing pairsn. LetSbe a set of conjugation pairs such that if two pairs interlace, they commute. First, it- eratively cancel all pairsU j,U † j such that no operation inSappears betweenU j andU † j . Forn= 0 this yieldsQ ...
[21]

S. Lamm, C. Schulz, D. Strash, R. Williger, and H. Zhang, Exactly solving the maximum weight inde- pendent set problem on large real-world graphs (2018), arXiv:1810.10834 [cs.DS]

2018 arXiv
[22]

Berman and T

P. Berman and T. Fujito, On approximation properties of the independent set problem for degree 3 graphs, in Workshop on Algorithms and Data Structures(Springer,
[23]

Sakai, M

S. Sakai, M. Togasaki, and K. Yamazaki, A note on greedy algorithms for the maximum weighted indepen- dent set problem, Discrete Applied Mathematics126, 313 (2003)

2003
[24]

M. Xiao, S. Huang, Y. Zhou, and B. Ding, Efficient re- ductions and a fast algorithm of maximum weighted inde- pendent set, inProceedings of the Web Conference 2021, WWW ’21 (Association for Computing Machinery, New York, NY, USA, 2021) p. 3930–3940

2021
[25]

Besides, of course, repeated operationsU i =U j and op- eration inversesU i =U † j
[26]

Grover and T

L. Grover and T. Rudolph, Creating superpositions that correspond to efficiently integrable probability distribu- tions (2002), arXiv:quant-ph/0208112 [quant-ph]

2002 arXiv
[27]

M. P. Fisher, V. Khemani, A. Nahum, and S. Vijay, Random quantum circuits, Annual Review of Condensed Matter Physics14, 335 (2023)

2023
[28]

Javadi-Abhari, M

A. Javadi-Abhari, M. Treinish, K. Krsulich, C. J. Wood, J. Lishman, J. Gacon, S. Martiel, P. D. Na- tion, L. S. Bishop, A. W. Cross, B. R. Johnson, and J. M. Gambetta, Quantum computing with qiskit (2024), arXiv:2405.08810 [quant-ph]

2024 arXiv
[29]

T. H. Cormen, C. E. Leiserson, R. L. Rivest, and C. Stein, Introduction to Algorithms, 2nd ed. (The MIT Press, 2001)

2001

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [18]

Naveh, S

A. Naveh, S. Ur, P. Emanuel, R. Alon, O. Kirzner, and E. Cornfeld, Automatic quantum circuit control skips (2024)

work page 2024

[2] [1]

M. A. Nielsen and I. L. Chuang,Quantum computation and quantum information(Cambridge university press, 2010)

2010

[3] [2]

N. P. de Leon, K. M. Itoh, D. Kim, K. K. Mehta, T. E. Northup, H. Paik, B. S. Palmer, N. Samarth, S. Sangtawesin, and D. W. Steuerman, Materials challenges and opportunities for quantum com- puting hardware, Science372, eabb2823 (2021), https://www.science.org/doi/pdf/10.1126/science.abb2823

doi:10.1126/science.abb2823 2021

[4] [3]

Krantz, M

P. Krantz, M. Kjaergaard, F. Yan, T. P. Orlando, S. Gus- tavsson, and W. D. Oliver, A quantum engineer’s guide to superconducting qubits, Applied Physics Reviews6, 10.1063/1.5089550 (2019)

doi:10.1063/1.5089550 2019

[5] [4]

C. D. Bruzewicz, J. Chiaverini, R. McConnell, and J. M. Sage, Trapped-ion quantum computing: Progress and challenges, Applied Physics Reviews6, 10.1063/1.5088164 (2019)

doi:10.1063/1.5088164 2019

[6] [5]

Henriet, L

L. Henriet, L. Beguin, A. Signoles, T. Lahaye, A. Browaeys, G.-O. Reymond, and C. Jurczak, Quantum computing with neutral atoms, Quantum4, 327 (2020)

2020

[7] [6]

Gidney, N

C. Gidney, N. Shutty, and C. Jones, Magic state culti- vation: growing t states as cheap as cnot gates (2024), arXiv:2409.17595 [quant-ph]

arXiv 2024

[8] [7]

Maslov, Advantages of using relative-phase toffoli gates with an application to multiple control toffoli opti- mization, Phys

D. Maslov, Advantages of using relative-phase toffoli gates with an application to multiple control toffoli opti- mization, Phys. Rev. A93, 022311 (2016)

2016